{"id":"d6320bd9-d29f-494b-9119-8bc092f56da9","arxiv_id":"2412.10180","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A reachability-based safety shield classifies possible human-robot contacts as constrained or unconstrained and limits robot kinetic energy to injury thresholds per contact type, enabling faster collaborative manipulation.","lead":"SARA shield is a safety layer for robot arms that uses reachability analysis to decide whether an upcoming contact with a person would be a clamping contact or a free-moving impact, and limits the robot's kinetic energy accordingly. It promises faster human-robot collaboration than existing safety methods while keeping contact energy below published pain and injury thresholds.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Energy verification checks only interval endpoints, so the formal guarantee in Eq. (10)/Eq. (6d) can be violated by an interior kinetic-energy peak.","rationale":"The reader's verdict is CONDITIONAL and I agree with that overall assessment, but I identify a different primary weakness. The reader highlighted the no-active-harmful-motion assumption in Sec. III-D as the most fragile load-bearing premise; that is an explicit modeling assumption shared with much of the power-and-force-limiting literature, and while untested, it does not by itself make the internal argument inconsistent. The more load-bearing concern is internal to the formal verification: Eqs. (47)-(48) check energy only at interval endpoints, yet Eq. (6d) requires an interval-wide bound. Under the paper's own definitions, the shield can certify a trajectory as safe even though the robot's kinetic energy exceeds the threshold at an interior time when a contact is possible. This is a soundness gap in the formal core, not merely an untested environmental assumption. The pendulum experiment does not repair it, because the measured pendulum energy is a lower bound on the transferred energy rather than an upper bound on the robot's pre-contact kinetic energy. The proposed test would settle the issue by checking whether a dense time-discretized energy maximum can exceed the endpoint maximum in a scenario the shield accepts. If the test fails (i.e., the endpoint check is sufficient in practice for the considered controller and time step), the concern would be resolved and the formal claim restored. The paper has real strengths: the reachability-based contact classification is novel, the efficiency results are strong, and the constrained-contact force measurements directly support part of H2. My recommendation remains CONDITIONAL, matching the reader's verdict, but with the condition refocused on replacing or rigorously justifying the endpoint-only energy verification.","tokens_in":25093,"tokens_out":3938,"duration_ms":44773,"concrete_test":"Construct a one-link or two-link trajectory where the intended controller accelerates to a high joint speed in the interior of [ta, tb] and then brakes so that q(ta) and q(tb) are both slow. Place a human reachable occupancy that intersects the robot only during the high-speed interior sub-interval. Run the SARA shield verification code and record csafe from Eq. (10) with the endpoint checks of Eqs. (47)-(48). Then densely sample T^r_i(q(t)) (e.g., 100 subdivisions of [ta, tb]) and compare the sampled maximum with the endpoint maximum. If csafe=true while the sampled maximum exceeds T_{type,i,j}, the formal guarantee is unsound. A minimal analytic version: for a 1-DOF joint with constant inertia, let qdot(t) be a triangular profile peaking at the interval midpoint; the endpoint check sees only the low endpoint speeds and misses the peak.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central formal claim is that whenever a contact is possible during a monitored trajectory, the robot's kinetic energy stays below the relevant threshold. Eq. (6d) requires this for every time t in the interval. However, Eqs. (47)-(48) define the energy checks cT,free and cT,clamp using only max(T^r_i(q(ta)), T^r_i(q(tb))) for each interval [ta, tb]. This endpoint maximum is not an upper bound on T^r_i over the whole interval. The effective kinetic energy T^r_i(q) = 1/2 qdot^T E_i B(q) E_i qdot is a configuration-dependent quadratic form; it can attain an interior maximum even for a smooth joint-space trajectory when the robot accelerates and then brakes within [ta, tb], or when the inertia matrix B(q) varies along the path. The monitored trajectory is exactly a concatenation of an intended trajectory (which may accelerate to minimize time) and a failsafe braking trajectory, so non-monotonic energy profiles are expected. Thus csafe from Eq. (10) can evaluate to true while a contact at an interior time would involve kinetic energy above the allowed threshold. This is an under-approximation in the wrong direction for a safety proof. The pendulum experiment in Sec. VII-B2 does not close this gap: the measured pendulum potential energy is a lower bound on the energy transferred from the robot, not an upper bound on the robot's kinetic energy at contact, so it cannot validate the endpoint check.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents SaRA shield, a power and force limiting framework for autonomous manipulators in human environments. It combines reachability analysis with contact-type classification: potential human-robot contacts are detected over a monitored trajectory, classified as constrained (clamping) or unconstrained, and the robot's effective kinetic energy at a contact candidate is verified against body-part-specific injury thresholds taken from ISO standards and recent surrogate studies. If verification succeeds the intended trajectory is executed; otherwise a failsafe braking trajectory is used. The authors report that SaRA shield significantly improves task efficiency over several baseline methods in simulation and on a real manipulator, that constrained-contact force measurements with a Pilz system respect the derived force limits, and that pendulum-collision experiments show the transferred energy remains below the set unconstrained-contact thresholds.","tokens_in":25300,"tokens_out":6966,"duration_ms":73796,"significance":"The reachability-based classification of constrained versus unconstrained contacts is a novel and practically valuable idea, and the formal development in Sec. V (Propositions V.1, V.2, V.3 and the combined-body-part proof in Appendix B) is detailed and mostly rigorous. The constrained-contact force validation with a certified Pilz measurement system is a concrete, reproducible test that supports H2 for clamped contacts. However, the central claim of a formal energy safety guarantee is currently undermined by the endpoint-only energy check in Sec. VI and by the use of the nominal trajectory rather than the reachable set for energy verification; the unconstrained-contact pendulum experiment also does not directly bound the robot's kinetic energy at contact. With those gaps closed, the framework would represent a meaningful advance for safe, non-conservative human-robot collaboration.","major_comments":[{"comment":"The energy constraints cT,free,i,j and cT,clamp,i,j are evaluated only at the interval endpoints q(ta) and q(tb), but the safety specification in Eq. (6d) requires the energy bound to hold for every t in the interval. The effective kinetic energy T^r_i(q) is a configuration-dependent quadratic form in the joint velocities and can attain an interior maximum even on a smooth trajectory, e.g., when the intended trajectory accelerates and the failsafe trajectory brakes within the same interval, or when the inertia matrix varies along the path. The checker in Eq. (10) can therefore return true while a contact at an interior time would involve kinetic energy above the admissible threshold. This is an under-approximation in the wrong direction for a safety proof. The paper needs an upper bound on T^r_i over each interval, derived from the reachable state set, not just the two endpoint states.","section":"Sec. VI, Eqs. (47)-(48)"},{"comment":"The energy verification uses the nominal monitored trajectory q(·) only, but the problem statement in Eq. (6d) restricts the 'maximal reachable kinetic energy' T^r_i(t) and accounts for the disturbance model W in the reachable occupancy. The actual robot state can deviate from the nominal trajectory within the assumed tracking-error bounds, and the kinetic energy of the deviating state can be larger than the nominal value. Since no disturbance set is propagated through Eqs. (47)-(48), the formal guarantee does not hold under the stated uncertainty model. The energy check should be applied to the reachable set of states over the interval, analogous to the occupancy verification.","section":"Sec. VI, Eqs. (47)-(48) and Sec. III-C"},{"comment":"The pendulum experiment measures the maximum potential energy of the pendulum after impact, which is a lower bound on the energy transferred from the robot, not an upper bound on the robot's kinetic energy at contact. The statement that 'the kinetic energy of the robot at the time of contact can never exceed the potential energy of the pendulum' is incorrect: the robot's pre-contact kinetic energy can exceed the pendulum's post-impact energy, with the remainder dissipated or retained in the robot. Consequently, the fact that the measured pendulum energy stays below the set threshold does not establish that the robot's kinetic energy at contact was below that threshold. The paper should either directly measure or compute T^r_i at the contact instant from joint states, or redesign the experiment to infer the robot's pre-contact energy by accounting for post-contact robot motion and energy losses.","section":"Sec. VII-B2, pendulum experiment"},{"comment":"The safety guarantee is conditional on the assumption that 'the human does not actively move into hurtful contact with the robot so that the speed difference between the human and the robot along the contact normal is always smaller or equal to the robot speed.' This assumption is not enforced or measured in the experiments, and it is in tension with the introductory claim of 'arbitrary human motion' in Sec. I. Since the energy thresholds in Sec. VI bound only the robot's kinetic energy, not the relative impact energy, a human moving toward the robot faster than the robot can violate the energy limit even when the robot itself satisfies Eq. (6d). The paper should state this conditionality prominently in the abstract and conclusions, and ideally provide an argument or measurement that the experimental scenarios satisfy the assumption.","section":"Sec. III-D, final assumption"}],"minor_comments":[{"comment":"In the definition of li, the text reads 'li = ∥pr_i,2 − pr_i,2∥2 + rr_i', which appears to be a typo; it should presumably be ∥pr_i,2 − pr_i,1∥2 + rr_i, matching the usage in Eq. (30d) and Appendix A.","section":"Sec. V-B2, Theorem V.3"},{"comment":"The sentence 'the kinetic energy of the robot at the time of contact can never exceed the potential energy of the pendulum' should be reworded to say that the energy transferred to the pendulum cannot exceed the robot's initial kinetic energy; as written it misstates the direction of the inequality.","section":"Sec. VII-B2, paragraph 2"},{"comment":"The claim of formal safety for 'arbitrary human motion' overstates the actual assumptions in Sec. III-D, which restrict human speed and assume no active movement into hurtful contact. Consider aligning the wording with the stated assumptions, e.g., 'under the stated bounded-human-motion assumptions'.","section":"Sec. I, contribution (c)"},{"comment":"The runtime comparison is informative, but the standard deviations are quite large for several methods (e.g., Reflected mass 1.00 ± 5.02 ms). It would be helpful to state the number of control cycles or trials over which the runtimes are averaged.","section":"Sec. VII-A, Table II"}],"recommendation":"major_revision","confidential_remarks":"The core gap is the endpoint-only energy check in Sec. VI, which is load-bearing for the paper's central formal claim. The good news is that the fix seems within scope: the authors already over-approximate reachable states and velocities for occupancy and clamping checks, and the same machinery could be used to bound the effective kinetic energy over each interval. The pendulum experiment is also fixable by measuring joint states at the contact instant. I would not reject the paper, but the formal guarantee as currently stated is not supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing you should know: this is a real contribution. Thumm et al. are the first to use reachability analysis to classify possible contacts as constrained or unconstrained, and to enforce contact-type-dependent energy thresholds. That distinction matters because unconstrained contacts tolerate much higher energy, and the efficiency numbers (92.6% vs. 38.5–64.5% for the main baselines) are convincing. The constrained-contact force measurements with the Pilz system support the clamping claim across edge, wedge, sheet, and blunt geometries, and the formal derivations in Sec. V and Appendix A are careful. The paper is honest about its assumptions, and the self-citations are not doing load-bearing work.\n\nNow the soft spots, in order of severity.\n\nFirst, the formal guarantee in Eq. (10)/(6d) is not established. The check cT in (47)–(48) only compares kinetic energy at the two endpoints of each interval. But T_i^r(q) is a configuration-dependent quadratic form, and the monitored trajectory is an intended trajectory followed by a braking failsafe. Nothing prevents an interior kinetic energy peak above the threshold, so csafe can be true while a contact at an interior time violates the energy bound. This is an under-approximation in the wrong direction for a safety proof. The verification needs to bound T over the whole interval, not just at endpoints.\n\nSecond, the pendulum experiment in Sec. VII-B2 uses an inverted energy argument. The measured pendulum potential energy is a lower bound on the energy transferred from the robot, not an upper bound on the robot’s kinetic energy at contact. Plastic deformation, friction, or sensor-stop timing can all absorb energy, so the experiment does not validate that robot kinetic energy stays below the unconstrained thresholds. The result is suggestive but not the claimed confirmation of H2 for unconstrained contacts.\n\nThird, the assumption in Sec. III-D that “the human does not actively move into hurtful contact” is load-bearing and untested. It is shared with many power-and-force-limiting approaches, but the abstract and introduction promise “arbitrary human motion,” which oversells. The speed-difference condition along the contact normal should at least be monitored, not just assumed.\n\nThese gaps are fixable. The endpoint issue needs a real interval-wide energy bound or a different verification invariant. The pendulum experiment needs a direct upper bound on robot energy, such as measured robot velocity at contact. The human-motion assumption can be stated more carefully or checked online.\n\nWho this is for: anyone working on safe human-robot collaboration, formal safety shields, or power-and-force limiting will get real value. The paper deserves a serious referee: accept with major revision, conditional on closing the formal gap. I would not desk-reject it.","headline":"The contact-type classification is genuinely new and the efficiency gains look real, but the formal energy guarantee only checks interval endpoints and the pendulum experiment cannot support the unconstrained-contact claim; both are fixable.","tokens_in":25869,"tokens_out":3122,"would_cite":true,"duration_ms":33651,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"SARA shield is a formal safety framework that classifies every potential human-robot contact as clamped or free, verifies the robot's impact energy against pain and injury thresholds for that contact type, and lets autonomous manipulators…","keywords":["human-robot collaboration","power and force limiting","reachability analysis","contact type classification","clamping detection","contact energy thresholds","formal verification","autonomous manipulation"],"falsifier":"Repeat the paper's pendulum collision test with the pendulum actively swinging toward the robot so that the relative speed along the contact normal at impact exceeds the robot's own speed: if the energy transferred to the pendulum exceeds the shield's allowed threshold for a trajectory the shield cleared, the formal claim as stated is falsified. A cheaper check is to measure the relative approach speed along the contact normal in real deployments and find it exceeding the robot speed while the shield remains active.","tokens_in":24841,"feed_emoji":"🤖","tokens_out":7125,"duration_ms":72371,"temperature":0.7,"pith_summary":"The paper is trying to establish that an autonomous robot arm can work at high speed in a shared human workspace and still carry a formal safety guarantee, provided the safety layer distinguishes between the two kinds of contact a person can have with it. The distinction matters because an unconstrained contact, where the human can pull away, tolerates 5 to 20 times higher impact velocities than a constrained (clamping) contact, where a body part is pressed against the robot or a surface. SARA shield uses reachability analysis to over-approximate everything the human and robot could do in the near future, detects every contact that could occur, classifies each as constrained or unconstrained, and slows the robot only when the classified contact would exceed the energy limit for that body part. The experiments show the robot completing 92.6 percent of its trajectory at full speed in close collaboration, versus 59.2 percent for the strongest baseline, while measured contact forces stay below admissible limits. If true, this gives collaborative robots a way to be both fast and certifiably safe.","feed_headline":"New safety shield keeps robot arms at 93% of full speed","feed_subtitle":"Reachability analysis tells clamping from free contact, so the robot slows only for genuinely dangerous touches.","key_machinery":"The load-bearing object is the monitored trajectory, a concatenation of the intended trajectory and a path-consistent failsafe braking trajectory, whose safety is checked before any action is executed. Reachability analysis over-approximates the occupancies of all human body parts and robot links over the prediction horizon, given a bounded human speed, a bounded measurement error and delay, and a bounded tracking error. The contact classifier then marks a potential contact as unconstrained only if no environmental clamping (human occupancy intersecting a static obstacle) and no self-clamping (human occupancy intersecting two robot links) is possible, with three relaxations: the gap between robot and obstacle must be smaller than the body-part diameter, the robot's velocity must actually point toward the obstacle, and known safe link topologies are exempted. On the energy side, the paper proves that the effective kinetic energy exerted through a contact on link $i$ equals $T_i^r(q) = \\frac{1}{2}\\dot{q}^{\\mathsf{T}} E_i B(q) E_i \\dot{q}$, independent of the rotational velocities of joints distal to $i$, which keeps the verification feasible at control frequency. Verified trajectories are then checked against per-body-part energy thresholds for blunt, wedge, edge, and sheet geometries, separately for constrained and unconstrained contacts.","core_discovery":"The central claim is that in each control cycle, SARA shield detects all possible contacts between the robot and the human using reachability analysis, categorizes them as constrained or unconstrained, and verifies that the kinetic energy of the robot link at contact is below the pain and injury thresholds for the detected contact type and body part. Because the verification is carried out on a monitored trajectory that consists of the intended motion followed by a failsafe braking trajectory, an unsafe action is never executed: the shield substitutes the last verified failsafe motion instead. The energy bound is computed from the robot's effective kinetic energy with respect to the contacting link, and the contact classification is made formally correct by checking occupancy intersections with the environment, the possibility of clamping between two robot links, the human body-part diameter, and the direction of robot motion. Under the stated assumptions, the paper claims this yields provable safety for dynamically changing robot paths, arbitrary human motion, and sharp robot geometries, with significantly less conservatism than speed-and-separation monitoring or worst-case power-and-force limiting.","pith_inferences":["The safety argument is one-sided in a specific sense: the energy bounds are stated in terms of the robot's kinetic energy, and the assumption that the human never approaches faster than the robot along the contact normal is not enforced or measured. A deployment that logs the actual relative approach speed at each contact would show whether the guarantee holds outside that assumption.","The same classification machinery should transfer to mobile manipulators and humanoid robots, which the paper names as future work, but the velocity-direction test for clamping assumes the occluding obstacle is static; a moving base or moving workpiece would require the signed-distance condition to be re-derived in relative coordinates.","The unconstrained-contact thresholds come from surrogate experiments with pendulums and artificial tissue, so the framework's practical safety margin inherits the fidelity of those data sets; re-running the verification with live-tissue injury data would be the natural way to tighten or relax Table I.","A testable consequence of the framework is that a robot executing a shield-cleared trajectory never delivers more than the tabulated energy to a passive human contactor; this could be monitored continuously in deployment with a force-torque sensor at the contact point, turning the formal bound into a runtime check."],"forward_implications":["Autonomous manipulators can run at near full speed through most of a collaborative task: SARA shield achieves 92.6 percent of the unshielded trajectory length in simulation and 93.7 percent on a real setup, versus 55.9 percent for dynamic speed-and-separation monitoring.","A robot controlled by SARA shield satisfies the contact-energy constraints for every contact type: measured constrained-contact forces stayed below the admissible limits for all five body parts and four end-effector geometries, and measured pendulum energies stayed below all five unconstrained thresholds.","Sharp end-effector geometries no longer force a blanket speed reduction: because most contacts are unconstrained, edge, wedge, and sheet tools permit nearly the same efficiency as blunt ones.","The framework is certifiable rather than heuristic: time delays, measurement errors, bounded human speed, and tracking errors are explicitly absorbed into the reachable sets, so the safety guarantee holds by induction over control cycles.","The contact-type classification itself, not the energy verification, is what unlocks the performance gain: without the $c_{\\mathrm{free}}$ term the approach performs no better than the reflected-mass baseline."],"supporting_citations":[{"why":"Supplies the safety-shield architecture of intended plus failsafe monitored trajectories that SARA builds on.","marker":"[19]"},{"why":"Provides the set-based human pose prediction and capsule reachable-occupancy computation used in every control cycle.","marker":"[24]"},{"why":"Defines the operational-space kinetic energy matrix used to compute the effective energy at the contacting link.","marker":"[27]"},{"why":"Contributes the surrogate experiments showing unconstrained contacts allow 5 to 20 times higher impact velocities than constrained ones.","marker":"[16]"},{"why":"Provides the ex-vitro injury data for edged, wedge, and sheet contact geometries that set the sharp-geometry thresholds.","marker":"[40]"},{"why":"Sets the ISO constrained-contact energy limits and the formula converting energy into an allowable transient contact force.","marker":"[17]"},{"why":"Establishes the reflected-mass baseline and the injury-database-to-safe-motion-unit methodology that SARA extends.","marker":"[11]"},{"why":"Is the dynamic speed-and-separation baseline that SARA is compared against and the origin of the online verification style.","marker":"[12]"},{"why":"Defines the 1.6 m/s human approach speed used to bound the human reachable sets.","marker":"[43]"}],"fun_headline_variants":["SARA shield: fast robot motion with formal contact safety","Reachability-based shield keeps robots fast while guaranteeing safety","Formal proof of safety lets robots work at full speed near humans","Contact-aware shield reduces slowdowns without compromising safety","Reachability analysis classifies contacts to keep robot energy safe"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole guarantee rests on the assumption that a person never moves toward the robot faster than the robot moves toward them along the line of contact; if someone actively lunges into or grabs the robot, the relative impact energy can exceed the amount the thresholds are designed to bound.","fun_headline_variants_meta":{"raw":{"variants":["SARA shield: fast robot motion with formal contact safety","Reachability-based shield keeps robots fast while guaranteeing safety","Formal proof of safety lets robots work at full speed near humans","Contact-aware shield reduces slowdowns without compromising safety","Reachability analysis classifies contacts to keep robot energy safe"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001764,"raw_usage":{"total_tokens":6959,"prompt_tokens":939,"completion_tokens":6020,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":555,"completion_tokens_details":{"reasoning_tokens":5938}},"tokens_in":555,"tokens_out":6020,"duration_ms":43463,"temperature":1.0,"reasoning_tokens":5938,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T16:16:12.728716+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the paper's pendulum collision test with the pendulum actively swinging toward the robot so that the relative speed along the contact normal at impact exceeds the robot's own speed: if the energy transferred to the pendulum exceeds the shield's allowed threshold for a trajectory the shield cleared, the formal claim as stated is falsified. A cheaper check is to measure the relative approach speed along the contact normal in real deployments and find it exceeding the robot speed while the shield remains active.","supporting_citations":[{"cited_title":"Provably safe deep reinforce- ment learning for robotic manipulation in human envi- ronments,","cited_arxiv_id":null,"evidence_quote":"Supplies the safety-shield architecture of intended plus failsafe monitored trajectories that SARA builds on."},{"cited_title":"SaRA: A tool for safe human–robot coexistence and collabora- tion through reachability analysis,","cited_arxiv_id":null,"evidence_quote":"Provides the set-based human pose prediction and capsule reachable-occupancy computation used in every control cycle."},{"cited_title":"Inertial properties in robotic manipulation: An object-level framework,","cited_arxiv_id":null,"evidence_quote":"Defines the operational-space kinetic energy matrix used to compute the effective energy at the contacting link."},{"cited_title":"Towards unconstrained collision injury protection data sets: Initial surrogate ex- periments for the human hand,","cited_arxiv_id":null,"evidence_quote":"Contributes the surrogate experiments showing unconstrained contacts allow 5 to 20 times higher impact velocities than constrained ones."},{"cited_title":"Towards safe robot use with edged or pointed objects: A surrogate study assembling a human hand injury protection database,","cited_arxiv_id":null,"evidence_quote":"Provides the ex-vitro injury data for edged, wedge, and sheet contact geometries that set the sharp-geometry thresholds."},{"cited_title":"Robotics - safety requirements - part 2: Industrial robot applications and robot cells,","cited_arxiv_id":null,"evidence_quote":"Sets the ISO constrained-contact energy limits and the formula converting energy into an allowable transient contact force."},{"cited_title":"On making robots understand safety: Embedding injury knowledge into control,","cited_arxiv_id":null,"evidence_quote":"Establishes the reflected-mass baseline and the injury-database-to-safe-motion-unit methodology that SARA extends."},{"cited_title":"Online verifica- tion of multiple safety criteria for a robot trajectory,","cited_arxiv_id":null,"evidence_quote":"Is the dynamic speed-and-separation baseline that SARA is compared against and the origin of the online verification style."},{"cited_title":"Safety of machinery - positioning of safeguards with respect to the approach speeds of parts of the human body,","cited_arxiv_id":null,"evidence_quote":"Defines the 1.6 m/s human approach speed used to bound the human reachable sets."}],"review_version":1}